Suppose that uN is of finite trace. Then there exist a set X⊂X whose complement X −X is μ-negligible and a function H∈L2(X×X) satisfying the following conditions :
(i) For every (x,y)∈X×X, the family (αifi(x)gi(y))i∈I is summable in C and its sum is N(x,y);
(ii) For every finite subset J of I and every (x,y)∈X×X, one has
∣i∈J∑αihi(x,y)∣⩽H(x,y)(14)
(iii) The function x↦H(x,x) belongs to L1(X).
Since uN is of finite trace, the family (αi) is summable (cor. 4 of IV, p. 165). The series
i∈I∑αi∣fi∣2,i∈I∑αi∣gi∣2
therefore converge in L1(X) and consequently μ-almost everywhere (INT, IV, p. 128, § 3, no3, prop. 6). Let F and G denote, respectively, the functions defined μ-almost everywhere by the sum of these series, putting F(x)=0 and G(x)=0 for every x such that the corresponding series does not converge. Since F and G belong to L1(X), the function H defined by H(x,y)=√F(x)G(y) belongs to L2(X×X) (INT, V, p. 95, § 8, no3, cor. 2).
Let us apply prop. 20 to the space X, to F=C2, and to the measurable mappings
sn:x↦(i∈I∑αi∣fi(x)∣2,i∈I∑αi∣gi(x)∣2)
i⩽ni⩽n
defined on X. There therefore exists a sequence (Cm)m∈N of compact subsets of X satisfying the following conditions:
(1) the union X satisfies μ(X−X)=0;
(2) for every m∈N and every n∈N, the functions sn are continuous on Cm;
(3) for every m∈N, the series ∑αi∣fi∣2 and ∑αi∣gi∣2 converge uniformly on Cm to F and G, respectively;
(4) the support of the measure induced by μ on Cm is equal to Cm (replacing Cm by the support of this measure if necessary).
Let us prove that the set X and the function H satisfy conditions (i), (ii), and (iii).
Let (x,y)∈X×X. For every finite subset J⊂I, one has
By inequality (15), the series ∑iαihi converges uniformly on Km×Kn for every (m,n)∈N2. Let N be the function on X×X defined by
N(x,y)=i∈I∑αihi(x,y)=i∈I∑αifi(x)gi(y)
for every (x,y)∈X×X and by N(x,y)=0 otherwise. The function N is measurable (INT, IV, p. 175, § 5, no4, th. 2), and it is continuous on Km×Kn for every (m,n)∈N2.
From (16), we have ∣N(x,y)∣⩽H(x,y) for all (x,y)∈X×X, and in particular N belongs to L2(X×X) (INT, IV, p. 84, § 5, no6, th. 5).
Let us prove that N=N on X×X, which will establish property (i). Let f and g be elements of L2(X). We have
⟨g∣uN(f)⟩=∫X×XN(x,y)f(x)g(y)d(μ⊗μ)(x,y)
(formula (12) of IV, p. 172). For all (x,y)∈X×X and every finite subset J of I, we have
∣i∈J∑αifi(x)gi(y)f(x)g(y)∣⩽∣f(x)g(y)∣H(x,y)
by formula (16). Since the right-hand side of this inequality is integrable on X×X (INT, V, p. 95, § 8, no3, cor. 2), we may apply Lebesgue's theorem (INT, IV, p. 137, § 3, no7, th. 6) and formula (13), and deduce that
Consequently, we obtain uN=uN, whence N=N in L2(X×X) (prop. 3 of III, p. 28, b)).
For all (m,n)∈N2, the functions N and N are continuous on Km×Kn and therefore are equal on Km×Kn (prop. 9 of INT, III, p. 69, § 2, no2), since the support of the measure induced by μ⊗μ on Km×Kn is equal to Km×Kn. Hence N coincides with N on X×X.
√
Finally, the estimate FG ⩽(F+G)/2 implies that the function x↦√F(x)G(x)=H(x,x) is integrable on X, whence property (iii).
In the rest of this number, we retain the notation of the proposition.
Bourbaki sets the proof directly after the statement, and the corpus does not separate them, so what is above is the statement and its proof together.